New paper out with FAIR(+FAIR-Chemistry):. Adjoint Sampling: Highly Scalable Diffusion Samplers via Adjoint Matching. We present a scalable method for sampling from unnormalized densities beyond classical force fields. 📄:

RT @guanhorng_liu: Adjoint-based diffusion samplers have simple & scalable objectives w/o impt weight complication. Like many, though, they….

RT @auhcheng: Excited to share Quetzal, a simple but scalable model for building 3D molecules atom-by-atom. 🐉 Named after Quetzalcoatl, th….

RT @RickyTQChen: Reward-driven algorithms for training dynamical generative models significantly lag behind their data-driven counterparts….

RT @RickyTQChen: We are presenting 3 orals and 1 spotlight at #ICLR2025 on two primary topics:. On generalizing the data-driven flow match….

This work was done during a PhD internship at FAIR NYC, thanks to my amazing supervisors @brandondamos ,.@RickyTQChen and Brian Karrer. Special thanks to our core contributors: @bkmi13.@bingyan4science @xiangfu_ml.@guanhorng_liu (and of course @cdomingoenrich).

Our evaluation offers a new challenging sampling benchmark for molecular conformer generation. The benchmark features real, drug-like molecules from the SPICE dataset, and we hope it drives direct and tangible progress in sampling for computational chemistry (coming soon).

This lets us train conditional diffusion samplers directly from expensive energy functions, namely, state-of-art molecular foundation models, amortizing sampling across thousands of molecules—unlike traditional samplers, which require heavy energy access per new molecule sample.

We specialize Adjoint Matching—originally designed for reward fine-tuning—to the sampling setting. By exploiting a factorization of the optimal transition density (a Schrödinger bridge), our new loss enables heavy reuse of simulations and energy evaluations.

RT @RickyTQChen: Want to learn continuous & discrete Flow Matching? We've just released:. 📙 A guide covering Flow Matching basics & advance….

Hi friends. I will be at the NYC FAIR office for the next 6 months as a research scientist intern. I’ll be working on all things control theory x generative modeling under the amazing @brandondamos. Please feel free to reach out if you’re around NYC!.

A virtual session on the intersection of optimization, control and RL happening now (9am-12pm CT) for the annual CSL student conference! Find the schedule and zoom link here:

RT @angelaschoellig: Our review paper on **Safe Learning in Robotics** is out, including open-source code. Wondering how model-driven and d….

(6/6) encapsulated the key perspectives I would like to attempt to bridge going forward: Control, Learning and Optimization. Thanks to my great PhD advisor Prof. Bin Hu who shares these same curiosities.

(5/6) unstable and unconstrained methods will likely fail without a large data set and alternative LQR-based constraint methods will be too biased to solutions with better margins. This was a rather straight-forward project to start my PhD, but I felt that it really.

(4/6) state-space model and small gain theorem from robust control. You can even model against time-varying polytopic bounded uncertainties just by adding a few more LMI constraints to your problem. This formulation can be especially useful when your system is nearly.

(3/6) I think one the coolest aspects of this work is that it lets you design your imitation learning problem almost as if you were a control engineer. Not only can you guarantee stability but you can also bias your controller against all kinds of uncertainties using the extended

(2/6) We consider the problem of learning a controller of a known LTI system from demonstrations while constraining the solutions to be robustly stabilizing using some classic LMI conditions. You can efficiently obtain a stationary point to this problem using an ADMM formulation.

(1/6) Happy to share one of the first papers of my PhD which was done last fall: "On Imitation Learning of Linear Control Policies: Enforcing Stability and Robustness Constraints via LMI Conditions" :

RT @viditnanda: I love all of you very much, so please don't beat me up for this.